EP2828794A1

EP2828794A1 - Method and apparatus for evaluating results of gaze detection

Info

Publication number: EP2828794A1
Application number: EP13710873.4A
Authority: EP
Inventors: Denis Williams; Jan Hoffmann
Original assignee: SensoMotoric Instruments Gesellschaft fuer Innovative Sensorik mbH
Current assignee: SensoMotoric Instruments Gesellschaft fuer Innovative Sensorik mbH
Priority date: 2012-03-22
Filing date: 2013-03-21
Publication date: 2015-01-28
Anticipated expiration: 2033-03-21
Also published as: CN104321785A; JP6067093B2; US9639745B2; US20150063635A1; EP3168783B1; JP2015514251A; EP2828794B1; CN104321785B; IN2014DN08740A; EP3168783A1; WO2013139919A1; EP2642425A1

Abstract

The invention relates to a method and an apparatus for evaluating results of gaze detection, wherein these results are present or are obtained in the form of information which defines, for each of a multiplicity of successive times, a viewing direction detected at this time and a focal point (12) identified thereby in a scene image (10) assigned to this time. For this purpose, the invention provides for the following steps to be carried out: - a temporal change in the viewing direction and/or the focal point (12) is evaluated in order to identify different viewing events which differ from one another by different speeds (v) of an eye movement, wherein saccades (14) and fixations (16) and/or pursuit movements (15) are detected as different types of viewing events and the identified viewing events are classified according to the type thereof, - a period of time spanned by the times is divided into intervals in such a manner that an interval corresponding to a duration of the particular viewing event is assigned to each of the identified viewing events, wherein at least some of these intervals each contain a sequence of a plurality of times, - precisely one of the times or a true subset of the times is selected in each case from each of the intervals (I₁ - I₆) assigned to a fixation (16) or a pursuit movement (15), - and, for each of these selected times, the focal point (12) identified in the scene image (10) assigned to the particular time is mapped to a position (13) corresponding to this focal point (12) in a reference image (11).

Description

Method and device for evaluating the results of a gaze detection

The present invention relates to a method and a device for evaluating the results of a gaze detection, that is to say the results of an eye tracking.

Eye tracking is known per se and is used, for example, in behavioral research or market research, when it has to be examined, in soft points of view, when or for how long and / or in which sequence or how the look or a line of sight moves between different points of view. This can, for example, give information on where and how a provider should place the goods offered at the lowest price in order to be sure of the attention of the traffic. In methods known from the prior art for eye tracking, scene images are recorded which reproduce a scene that is visible to a subject - the word may also refer in the following to both a person and an animal - and for which, for example, a can be used on a head of the subject attached and so moving with the head camera. At the same time an eye position and thus a viewing direction of the subject is detected, which can be done for example with the help of targeted to the eyes of the subject cameras using appropriate image analysis. After a calibration of such an eye tracker can then be closed even with changing eye positions and head positions each on a viewpoint in the scene images, the term point of view - this also applies to the following description of the invention and the embodiments - a point in the scene on which the subject's gaze rests at a time when the scene biid is assigned.

Thus, results of a gaze detection are typically present as information that defines for each of a multiplicity of successive points in time a viewing direction detected at that time and a viewpoint identified thereby in a scene image assigned to this point in time. In order to evaluate such results, a very large amount of information has to be processed because typically more than twenty scene images and pixels are captured per second. Regardless of whether the evaluation is done manually, automatically or interactively, it is therefore associated with a very large time or computational effort.

The present invention has for its object to propose measures with which the effort for the evaluation of results of a gaze detection of the described type can be reduced.

This object is achieved by a method having the features of the main claim and by a device with the features of the independent claim. Advantageous embodiments and further developments of the invention will become apparent with the features of the subclaims. The device according to the invention is suitable for carrying out the proposed method, which will be described first below, and which results in a significant reduction in the effort required to evaluate the results of a gaze detection. This method thus serves to evaluate the results of a gaze detection, wherein these results are present or obtained, for example in the manner described above by eye-tracking, as information which for each of a multiplicity of consecutive times respectively detects a viewing direction detected at that time and a thereby defining the viewpoint identified in a scene image assigned to this point in time, whereby the scene brothers are time-dependent, so that different scene images are assigned to the different points in time in the scene. This procedure comprises the following steps:

Evaluating a temporal change of the viewing direction and / or the viewpoint and / or an eye movement for identifying different visual events, wherein at least two different types of visual events are recognized and the identified visual events are classified according to their type,

Dividing a time period spanned by the times into intervals such that each of the identified gaze events is assigned an interval which corresponds to a duration of the respective gaze event, typically at least some of these intervals each containing a sequence of several of the times,

Selecting each of exactly one of the times or a true subset of the times from each of the intervals associated with a gaze event of at least one selected type of gaze event,

and, for each of these selected times, mapping the view point identified in the respective scene point in time, that is to say the point of view defined by the results of the gaze detection for that point in time, to a position corresponding to this viewpoint in a reference picture. The dividing of the period into the tiling intervals can take place, for example, by assigning to each of the times or scene images an attribute which indicates to which of the intervals or viewing events this time or the time at which this scene image is associated belongs. Typically, the intervals will be disjoint. The reference picture may be identical for the different scene pictures. But it is also possible that the reference image is one of a sequence of reference images, which may be defined by a reference video, for example. Similar to the scene images in this case, the reference image is time-dependent.

It is particularly expedient if, in particular, viewing events in which information is recorded and viewing events in which no information is recorded are recognized as different types of visual events - namely as visual events with information recording and as visual events without information recording. The above-mentioned gaze events of the at least one selected type of gaze event will then be the gaze events with information capture. The selection of exactly one of the times or of a subset of the zones is then generally effected, in particular, for each of the inverses assigned to a gaze event in which information is acquired.

The proposal to drastically reduce the effort for evaluation in this way is based on the following three insights. First, for many eye tracking applications alone or primarily, it is important to determine when a subject visually picks up what information. Secondly, at least one item of information relevant to the purposes of the proposed procedure does not take place only in certain types of gaze events, in particular fixations or follow-up movements in which the gaze rests for a long time more or less on an object resting or moving relative to the subject However, for example, in saccades, which represent a different kind of visual events and in which a too rapid eye movement makes a relevant in this sense information recording virtually impossible. Third, during the course of an information-related visual event, such as fixation or follow-up, the gaze always rests more or less on the same point or object, so that it suffices to examine from each fixation or sequential movement only for one time the scene image associated with that point in time and the viewpoint identified therein, in order to know everything necessary about the information recording during this viewing event. Incidentally, it is not absolutely necessary for the present invention described

Distinguish fixations and follow-up movements from each other. Conveniently, however, these are at least distinguished from events without information recording, such as saccades. Even taking into account this knowledge, however, it is not trivial to avoid unnecessary expenditure when evaluating the results of the eye tracking. This is only possible by evaluating the temporal change of the viewing direction and / or the viewpoint and / or the eye movement different visual events - preferably eye events with information recording and visual events without information recording, so for example. Saccades and fixations and / or sequential movements - are recognized as different types of gaze events and the period spanned by the times is divided into intervals such that each of the identified gaze events is assigned an interval corresponding to a duration of the respective gaze event. Only then can each of the intervals corresponding to a visual event of selected type - e.g. a viewing event with information recording or e.g. a fixation or a follow-up movement are assigned, in each case exactly one of the times or a real subset of the times are selected from this interval and be sure that an evaluation of those times assigned

Scenery pictures and points of view are sufficient without risking an unwanted loss of information.

It is not necessary to select more than one time from the individual intervals. However, the latter can be advantageous in order to eliminate or reduce errors or inaccuracies which may be unavoidable in view of the subsequently occurring and possibly automatically performed mapping into the reference image. In any case, the number of times selected and thus the mappings to be made of viewpoints in the reference image is significantly smaller than the number all times for which information about scene images and viewpoints is available.

The mapping of the viewpoints to corresponding positions in the reference image finally makes the systematic evaluation possible only because the positions determined here contain an objective statement as to what the subject's gaze has lain when and for how long, if the subject is to be allowed to do so Therefore, the scene images assigned to the different times are different from each other. A particular advantage arises from the fact that the number of images to be carried out by the proposed approach can be kept within very narrow limits. The method may also provide for analyzing the gaze behavior of several subjects in succession or simultaneously as described. The mapping of the viewpoints into the reference image, which is preferably chosen to be the same for all probands, then allows a very objective evaluation of the test results with very little effort.

It should be made clear that the terms used herein in connection with the identification and classification of visual events are "perceptions of a person taking an ingestion" or "visual events with an informational intake" on the one hand and "perceptions of a person in which no information is taken" or "visual events without On the other hand, information recording "on the other hand should not imply that in each case it is directly examined whether and to what extent the subject actually takes up information during the viewing event. What this means is that different visual events, which can usually be identified on the basis of typical physiological or motor features, are distinguished and that these visual events can be classified on the basis of empirical values or plausible considerations into those which are expected to receive information can, and those in which - eg due to excessive movement speed - it can be assumed that at least no relevant information is being collected. So you can, for example In this sense, it is assumed that information is taken during a fixation or a follow-up movement, but not at a saccade. Typical physiological

Characteristics in which the different types of visual events For example, they may be a speed of eye movement or a magnitude of an angle swept over a certain period of time. Finally, mentioning the visual events with information recording and the visual events without taking up information should not imply that precisely these and only these two types of visual events are distinguished. On the contrary, it is possible to identify a larger number of different types of visual events, which may also be assigned to a larger number of different classes, but which may each be interpreted in the sense mentioned as one of possibly different conceivable visual events. without information.

As can be seen from what has already been said, the different viewing events can differ from one another due to different speeds and / or amplitudes of an eye movement. Accordingly, the identification of the various viewing events can take place, for example, by detecting and evaluating a speed and / or amplitudes of an eye movement. In addition, further information can be evaluated for this purpose, eg via an eye acceleration or an image content in an environment of the viewpoints in the scene images, which can be subjected to a Btld evaluation, or via a head movement detectable with a head tracker. In particular, in expedient embodiments of the method, saccades will be recognized as gaze events in which no information is recorded, while fixations and / or subsequent movements can be recognized as gaze events in which information is acquired. The method can be configured such that for each of the intervals that are associated with a visual event of selected type-for example, a gaze event with information recording, for example, a fixation or a follow-up movement-a position in the reference image together with a reference image or "mapping end point" Information about the duration of each view event is stored. If exactly one of the times from this interval is selected, this position can be selected as the position tion that maps the viewpoint defined for the time selected from this interval. If more than one of the times are selected from this interval, that position may be selected as one of the positions or a-not necessarily arithmetic-mean or median of the positions to which the viewpoints defined for the times selected from that interval are mapped. In the latter case, the imaging target may be selected, for example, as the position to which the viewpoint is mapped, defined for the last time selected from this interval, or, in particular in the case of automatic mapping, in dependence on a measure of the quality of the respective image , In addition to the information about the duration of the respective gaze event, further attributes with information about the gaze event can also be stored together with the position, eg a start and an end, a consecutive number of the gaze event or a count for multiple occurrence of the same position. All of this information then allows a largely automatable evaluation of the results of the eye tracking.

An embodiment in which the method serves not only the evaluation of the gaze detection, but also the gaze detection itself, provides that the scene images are taken at the stated times with a camera, wherein the viewing directions are detected by means of an eye tracker, the Viewpoints in the scene biiders depending on the thus obtained information on the viewing directions determined. In this case, the camera can be worn by a head of the subject, the viewing directions are detected by the eye tracker. The eye tracker may e.g. Eye cameras for observation each have one of two eyes of the subject and provide an image analysis. The latter can e.g. determine a pupil position relative to a reflex on a cornea of the respective eye. For example, the two eye cameras may be held together with the former camera for taking the scene images from a spectacle-type carrier.

In the case of expedient embodiments of the described evaluation, the reference image is an image of a scene visible in the scene images, which corresponds to a fixed perspective and is photographically recorded, for example or sketching character, or an abstract rendering or coding of characteristic features recognizable in the scene creators. Such abstract rendering or coding may be given, for example, for each characteristic feature to be considered by a designation or location definition of that feature.

It can be provided that, for each of the selected points in time, the scene image associated with this point in time is output together with the viewpoint identified therein, e.g. on a screen. This can be done manually, e.g. with a mouse click, or to visually verify the visualization of the viewpoint in the reference image. The reference image is preferably also output for this purpose and may be e.g. next to or above or below the current scene image on the screen.

It should not be ruled out that the mapping of the pixels from the individually selected scene images into the reference image is performed manually. However, it is at least possible that the mapping of the viewpoints on the positions and thus the determination of said positions is carried out automatically by

Identifying a plurality of characteristic features and locations of these features in the reference image by means of a feature detection algorithm,

Identifying these features and determining locations of these features in the respective scene image by applying the same or a corresponding algorithm to the scene image,

Determining a transformation that maps the locations of the features in the scene image to the corresponding locations in the reference image, that is to say the locations of the respective features in the reference image, and

- Apply this transformation to the viewpoint identified in the respective scene image. The transformation can be determined, for example, under the constraint that it is a homography with the desired imaging properties. It can also be determined as a spatial image. In this case, at least some of the features may be located on the same scene object. The last three of the four steps mentioned here are carried out in each case for each of the selected times or each of the scenes assigned to these times. The characteristic features that are identified by the algorithm in the context of an image analysis may be, for example, corners or lines. Corresponding algorithms for feature detection are known per se, for example from the document US

6,711,293 bl.

Also, the selection of the times can be done both manually and automatically. So z, B. in each case a time is selected, which is defined by the fact that it is centered in the respective interval.

As can be seen from what has already been said, the proposed device is suitable for advantageously low-cost evaluation of results of a gaze detection, these results being present or readable or producible as information that for each of a plurality of successive times each detected at that time sight and define a view point identified thereby in a scene image assigned to this point in time, the scene bids being time-dependent, so that different scene images are generally assigned to the different points in time. For this purpose, this device is program-technically set up to carry out the following steps:

Evaluating a temporal change of the viewing direction and / or the viewpoint and / or an eye movement for identifying different visual events and recognizing at least two different types of visual events, and classifying the identified visual events according to their nature,

- dividing a time span spanned by the times into typically disjunctive intervals such that each of the identified gaze events an interval is assigned which corresponds to a duration of the respective gaze event,

Outputting the information defining the viewpoint in the scene image for selected points in time from the intervals associated with a view event of a selected type,

and, for each of the selected times, storing a position in a reference image as an image of the viewpoint defined for that point in time, and storing information associated with that position as an attribute over the duration of the view event or interval to which that point belongs; possibly together with further attributes with information about this gaze event, eg, about the beginning and end and / or a sequential number of the gaze event.

The device can be set up to divide the time period into the intervals by assigning to each time point in each case an attribute which indicates to which of the intervals or viewing events this time belongs. Again, the reference picture may be identical or time-dependent again for the different scene images.

Typically, the different types mentioned differ from one another in such a way that, in particular, viewing events with and sight events without information recording-as explained above-are distinguished from one another. The device is therefore set up in appropriate embodiments, viewing events in which an information recording takes place, and viewing events in which no information is recorded, to recognize different types of visual events. Accordingly, the mentioned visual events of a selected kind are expediently the visual events with information recording.

In order to be able to differentiate between the various visual events and, in particular, as visual events in which information is recorded and as visual events in which no information is recorded, the device can be set up to view the different visual events as a function of different speeds and / or To identify amplitudes of an eye movement. In expedient embodiments, the device is in particular configured to recognize saccades as gaze events without information recording and / or fixations and / or subsequent movements as gaze events with information recording.

Embodiments of the device are possible in which this is set up by programming, the positions defined for the selected times automatically map corresponding positions in the reference image on these points of view and to store the positions thus determined as the images of the viewpoints. For this purpose, it can be set up to automatically perform the following steps in order to map the viewpoints to the corresponding positions in the reference image:

Determining a transformation that maps the locations of the features in the scene image onto the thus identified locations of the respective feature in the reference image, and

- Apply this transformation to the viewpoint identified in the respective scene image.

For example, the transformation can be a homography. So that the viewpoints are imaged in the desired manner in the reference image, the device can thus be set up to determine the transformation in each case under the constraint that it is a homography with the required imaging properties with respect to the locations of said plurality of characteristic features. It is also conceivable that the device requires a manual input to define the mapping of the viewpoints from the selected scene images or allows correction as an automatically proposed image. In this case, the device can be set up for each of the selected times the corresponding position in the reference image together with the attribute assigned to these positions, depending in each case on an input defining the position and linking it to the time or the scene image assigned to this time to save. This input can be made, for example, for each of the selected times by a mouse click in the reference image shown to it.

Also for the selection of the times different possibilities are conceivable. Thus, the device can be set up for an input that defines the selected times as such - eg. For example, by a mouse click on a timeline on which the intervals are shown visibly displayed - or for an automatic selection of these times by applying these times as a function of the intervals defining rule. This requirement may e.g. define that the respective points in the middle of the intervals are selected.

The apparatus may also be arranged, if a plurality of times from the same interval associated with a view event with information recording is selected, to serve as an imaging destination associated with that interval one of the positions or an average of the positions to which the points of view belong Scene images are assigned, which are assigned to the selected from this interval times. The term mean here should not only be able to designate an arithmetic mean but also, for example, a median. It may be provided that the device further comprises an eye tracker for

Detecting the viewing directions and having a camera for taking the scene images, in this case is a device for viewing and evaluating the results of this view capture. The eye-tracker is then preferably provided with a holder for attachment to or on a head of the subject in order to allow a free movement of the subject in order to obtain as realistic information as possible about the subject's vision. habits can be obtained. The camera for capturing the scene images is typically part of the eye tracker and can be worn by the same holder. This facilitates calibration of the eye tracker.

Finally, the device can also have a camera for recording the reference image.

The method described above can be carried out in particular with a device of the type described here. On the other hand, in addition to its essential or optional features explained here, the device described can additionally be set up to carry out any desired embodiment or configuration of the method described at the outset.

An embodiment of the invention will be explained below with reference to Figures 1 to 6. It shows

1 is a schematic representation of a device for capturing and evaluating the results of this gaze detection,

2 shows a scene image taken with this device and, to the right of it, a reference image likewise recorded with this device, wherein an arrow illustrates how a viewpoint identified in the scene image is imaged onto a corresponding position in the reference image, FIG.

3 is a diagrammatic representation of a time profile of a speed of an eye movement detected with the device over an observation period and FIG

Fig. 4 is a timeline on which this observation period in several

Intervals are shown divided, the intervals corresponding to different with the device identified visual events. Thus, the device shown in FIG. 1 is an arrangement which comprises an eye tracker 1 for a gaze detection and is additionally set up, with this eye tracker 1 obtained results of Bficker- Constitution in a particularly appropriate manner with little effort to evaluate.

The eye tracker 1 has two eye cameras 2, which are held by a spectacle-like frame 3 and are directed to two eyes of a subject, not shown here, who carries the frame 3. In addition, the eye tracker 1 has a camera 4 held by the same frame 3 for taking scene images which is arranged to move with a head of the subject and that a field of view of this camera 4 corresponds to approximately one visual field of the subject , In addition, the eye tracker comprises a computing unit 5, which analyzes an output of the eye cameras 2 by means of image evaluation methods and thereby determines a current viewing direction of the test person for a plurality of closely spaced successive times. For this purpose, the arithmetic unit 5 may be e.g. be set up to determine a relative position of the pupils of the eyes to defined reflections on the cornea of the eyes, wherein the reflections can be generated by appropriately targeted illumination of the eyes. The camera 4 records a scene image assigned to this time at each of the named times. These scene images thus form successive frames of a film recorded with the camera 4. Depending on the particular viewing directions, the eye tracker 1 now identifies with the arithmetic unit 5, for each of the times, a viewpoint in the scene image assigned to this point in time, the viewpoint-also referred to as "gaze point" or "focal point" -in each case the point in the scene design, on which the subject looks at the given time.

In addition to the eye tracker 1, the device also has another camera 6, with which a reference image is taken. The reference image is a photographic image, taken from a defined perspective, of a scene which corresponds to the scene visible to the subject and which is also reproduced by the scene images. These Scene contains in the present case, a shelf 7, which may be, for example, a goods shelf in a supermarket and in which several objects 8, for example, different goods are placed. Embodiments are also conceivable in which the camera 6 is given by a video camera and the reference image is time-dependent.

Finally, the device has an evaluation unit 9 for evaluating the results obtained with the eye tracker 1. These results are information obtained in the manner described, which for each of the stated times defines the viewing direction detected at that time and the viewpoint which was identified in the scene image associated with this time.

In FIG. 2, one of the scene images 10 recorded with the camera 4 is shown on the left as an example, while the reference image 11 taken with the camera 6 is shown on the right next to it. Instead of this photographic reference image 11, in a modification, a sketch-like representation of the scene could also serve as a reference image or an abstract reproduction, in which recognizable characteristic features in the scene images - e.g. the items placed on the shelf 7 8 - are simply coded by their name or place. In the scene image 10, the viewpoint 12 identified in this scene image 10 is also represented in the form of an X.

For a systematic evaluation of the results of the gaze detection, 11 positions are now to be determined in the reference image, which the test person has looked at, and these positions should be stored together with information on when and for how long the test person looked there. In FIG. 2, an arrow illustrates how the viewpoint 12 is imaged onto a position 13 corresponding to this viewpoint 12. So that for each of the scene images such an image does not have to be performed in order to obtain the desired information, the procedure is now as described below.

The evaluation unit 9, which is set up in accordance with the program, detects a temporal change in the viewing direction and / or the viewpoint by comparing the tnforma that is present for the successive times. tions about the direction of vision and the point of view. FIG. 3 shows a time curve of a speed v of an eye movement of the test person within an observation period, with this observation period extending from to to ti. By comparing the respectively determined velocities v with a threshold value v _th different eye events are identified, ie saccades 14, in which the eyes move very fast, so that v> v _th , following movements 15, in which the eyes move much slower , so that v <v _th , and fixations 16, where the eyes stand still. The thus identified visual events are classified according to their type, in the example shown here in each case as a saccade

14, sequential movement 15 or fixation 16. Here, saccades 14 are recognized as a type of gaze event in which no information is captured, while sequential movements 15 and fixations 16 are recognized as two types of gaze events in which information is acquired. For the method described, however, it is not absolutely necessary to distinguish between subsequent movements 15 and fixations 16. These two types of gaze events can also be treated equally in the classification and can be easily classified as events in the gaze event with information gathering category.

By way of example, the preceding paragraph describes only a relatively simple method for identifying the various viewing events. In order to recognize the visual events of various kinds and to differentiate, in particular, as visual events with information recording and visual events without taking up information, additional measures may be provided, eg an evaluation of a head movement detected simultaneously with a head tracker or an IMU and / or an analysis of Image contents at the viewpoints 12 through an evaluation of scene camera data. As an alternative or in addition to the described evaluation of the speed v of the eye movement, an amplitude of eye movements can also be examined by comparing the viewing directions and / or viewpoints 12 at the different successive times and the respective visual event can be identified and classified depending on the detected amplitude , In this context, the amplitude is a maximum viewing angle difference observed during an eye movement within a certain period of time. For example, microsaccades, in which despite large Speeds with an information recording is expected to be distinguished from actual saccades, which can be treated as visual events without information recording. It is also conceivable to detect accelerations of the eyes additionally or instead for identifying the visual events.

After identifying and classifying the gaze events, the said observation period is divided into disjoint intervals by the evaluation unit 9, so that an interval corresponding to a duration of the respective gaze event is assigned to each of the identified gaze events. This is illustrated in Fig. 4, which shows a timeline with the intervals. The intervals which are assigned to a fixation 16 or a following movement 15 are designated there as intervals Ii, l ₂ ,,, and h. At least these intervals Ii to le each contain a plurality of said times. The division of the observation period into the intervals is done by assigning to each of the times an attribute indicating which of the intervals or gaze events this time belongs to.

Now, a time is selected for each of the intervals Ii to which are associated with a fixation 16 or a follow-up movement 15. The selected times are illustrated in Fig. 4 each by an arrow. Depending on the implementation, these times can either be selected automatically by applying a prescription which defines the selected time points in each case as times centrally located in the respective interval or in some way optimally for the algorithm used, or manually by clicking on the desired times on a this purpose on a screen 17 of the evaluation unit 9 reproduced timeline. In this case, the time beam can be reproduced in the manner shown in FIG. 4 in a manner which makes the relevant intervals I * to ₁₆ visible.

The evaluation unit 9 is set up such that for each of the selected points in time, the scene image 10 assigned to this time is output together with the viewpoint 12 identified therein on the screen 17, the reference image 11 each being shown in the manner shown in FIG displayed next to or above or below the scene image 10 on the screen 17.

For each of these selected times - and only for these times - the viewpoint 12, which is identified in the scene image 10 assigned to this point in time, is now mapped to the position 13 in the reference image 11 corresponding to this pixel 12. In this case, for each of the selected points in time, the position 13 in the reference image 11 which corresponds to the viewpoint 12 assigned to this point in time is represented as an image of the viewpoint 12 together with an attribute assigned to this position 13

Information about the visual event or about the interval to which this time belongs is stored. In particular, the duration of the respective viewing event or interval can be taken from this information. In addition, other attributes with information about this gaze event, e.g. Information about the beginning and end of the visual event, a continuous

Number of the gaze event or a count on multiple occurrence of the same position 13, each stored together with the position 13. It can be provided that the imaging of the viewpoints 12 from the selected scene images 11 manual is performed. This can e.g. each done by a mouse click, after an input mark or a cursor has been guided to the respective position 13 in the reproduced on the screen reference image. The evaluation unit 9 is then set up, for each of the selected times, the corresponding position 13 in the reference image 12 together with the attribute assigned to these positions 13 in each case depending on a position 13 defining the position 13 and associating it with the time or the scene image 10 assigned to that time Save input. The linkage may be e.g. This is done by the input defining the mapping always immediately following the input selecting the time.

However, the evaluation unit 9 can also be set up to automatically map the viewpoints 12 defined for the selected times to the positions 13 corresponding to these viewpoints 12 and to store the positions thus automatically determined as images of the viewpoints 12. The representation of the scene images 10 and the reference image 11 can then serve to visually check the automatic imaging of the viewpoint 12 into the reference image 11, even if the automatically determined position 13 is displayed there visibly.

For automatically mapping the viewpoints 12 to the corresponding positions 13 in the reference image 11, the evaluation unit 9 can be set up, a plurality of characteristic features, such as e.g. To identify corners or edges, and locations of these features in the reference image 11 by means of a feature detection algorithm and in addition to carry out the following steps for each of the selected times:

Identifying these characteristic features and determining locations of these features in the respective scene image 10 by applying the same algorithm to the scene image 10,

Determining a transformation which is a homography and images the locations of the features in the scene image 10 onto the so identified locations of the respective features in the reference image 11, and

Applying this transformation to the viewpoint 12 identified in the respective scene image 10.

Other than described above, it is also possible to select a plurality of timings for each of the intervals Ii to l ₆ or for some of these intervals Ii to l & min in order to minimize inaccuracies caused by erroneous mappings. In any case, significantly fewer times are selected than the respective interval contains. Of course, the term "time" here always refers only to times from the finite number of times for which information about points of view is available. If a plurality of points in time are selected from an interval for each of which a picture of a described type and thus a respective position 13 in the reference picture 11 is defined, one of these positions 13 or an average or median of these positions 13 can be stored together with the information associated with this position 13 as an attribute over the duration of the interval. The information thus determined and stored with comparatively little effort and therefore relatively quickly contains all the information needed for a further evaluation of the results of the gaze recognition and can be statistically evaluated in a variety of ways both mechanically and manually.

Claims

claims

Method for evaluating the results of a gaze detection, wherein these results are present or obtained as information which, for each of a multiplicity of consecutive times, identifies a viewing direction detected at that time and a viewpoint (12) identified thereby in a scene image (10) associated therewith. wherein the scene images are time dependent, such that the scene buffers associated with the different times are different, and wherein the method comprises the steps of;

Evaluating a temporal change of the viewing direction and / or the point of view (12) and / or an eye movement for identifying different visual events, wherein at least two different types of visual events are recognized and the identified visual events are classified according to their type,

- dividing a period spanned by the times into intervals such that each of the identified gaze events is assigned an interval which corresponds to a duration of the respective gaze event,

Selecting each of exactly one of the times or a true subset of the times from each of the intervals (Ii-1 ₆ ) associated with a gaze event of at least one selected type of gaze event,

and, for each of these selected times, mapping the viewpoint (12) identified in the respective scene at the respective point in time onto a position (13) corresponding to this viewpoint (12) in a reference image (11). A method according to claim 1, characterized in that when evaluating viewing events in which an information recording takes place, and viewing events in which no information recording takes place, are recognized as different types of Biickereignissen, wherein each selecting exactly one of the times or a subset of the times for each of the intervals (Ii-Ig) associated with an information-aware paging event occurs.

Method according to one of claims 1 or 2, characterized in that the different visual events differ from one another by different speeds (v) and / or amplitudes of eye movement and / or that saccades (14) are recognized as visual events in which no information is recorded, and / or that fixations (16) and / or subsequent movements (15) are recognized as visual events in which an information recording takes place.

Method according to one of claims 1 to 3, characterized in that for each of the intervals (Ii - 1 ₆ ) associated with a gaze event of selected type, a position (13) in the reference image (11) together with information about the Duration of each view event is stored, this position (13) is selected

as the position (13) on which the viewpoint (12) is mapped, which is defined for the point in time selected from this interval (Ii-Ig), if exactly one of the points in time from this interval (IIe) is selected, or

as one of the positions (13) or an average of the positions (13) on which the viewpoints (12) are mapped, which are defined for the times selected from this interval (Ii - 1 ₆ ), if several of the times are therefrom Interval (ii - \ $) are selected.

5. The method according to any one of claims 1 to 4, characterized in that the Szenierbifder (10) at said times with a camera (4) are recorded, the viewing directions are detected by means of an eye-tracker (1), the viewpoints (12) determined in the scene images (10) in dependence on the thus obtained information about the viewing directions, wherein the camera (4) is preferably carried by a head of a subject whose viewing directions are detected by the eye tracker (1).

6. Method according to claim 1, characterized in that the reference image (11) is an image of a scene that is visible in the scene images (10), which corresponds to a fixed perspective, or an abstract reproduction or coding of in the scene image (11). 10) recognizable characteristic features.

7. The method according to any one of claims 1 to 6, characterized in that for each of the selected times in each case assigned to this time scene image (10) is output for manually performing or visually checking the image of the viewpoint (12) in the reference image (11) ,

8. The method according to any one of claims 1 to 7, characterized in that the imaging of the viewpoints (12) on the positions (13) is carried out automatically by

identifying a plurality of characteristic features and locations of these features in the reference image (11) by means of a feature detection algorithm,

Identifying these features and determining locations of these features in the respective scene image (10) by applying the same or a corresponding algorithm to the scene image (10),

Determining a transformation that maps the locations of the features in the scene image (10) to the corresponding locations in the reference image (11), and Applying this transformation to the viewpoint (12) identified in the respective scene image (10).

9. A device for evaluating the results of a gaze detection, these results being present or being readable or producible as information which, for each of a multiplicity of consecutive times, respectively indicates a gaze direction detected at that time and a viewpoint (12) identified thereby at a time associated with this Defining scene image (10), wherein the scene images are time-dependent, such that the scene buffers associated with the different time points are different, and wherein the device is arranged to perform the following steps:

Evaluating a temporal change of the viewing direction and / or the viewpoint (12) and / or an eye movement for identifying different visual events and recognizing at least two different types of visual events, and classifying the identified visual events according to their nature,

Outputting the information defining the viewpoint (12) in the scene image (10) for selected points in time from the intervals (l _x - 1 ₆ ) which are associated with a look event of a selected type,

and, for each of the selected instants, storing a position (13) in a reference image (11) as an image of the viewpoint (12) defined for that point in time, and storing information of that position as attribute associated with that position (13) Duration of the visual event or interval (II) to which this time belongs Device according to claim 9, characterized in that it is arranged to recognize viewing events in which information is recorded and viewing events in which no information is recorded, as different types of visual events.

Device according to one of claims 9 or 10, characterized in that it is arranged to assign to the period in the intervals of each of the times each an attribute indicating which of the intervals or BItckereignisse this time belongs.

12. Device according to one of claims 9 to 11, characterized in that the device is set up to identify the various bending events as a function of different speeds (v) and / or amplitudes of an eye movement and / or to recognize saccades (14) as gaze events in which no information is recorded, and / or fixations (16) and / or subsequent movements (15) to be recognized as gaze events in which an information is recorded.

13. Device according to one of claims 9 to 12, characterized in that it is adapted to automatically represent the positions (13) in the reference image (11) corresponding to those points of view (12) defined on the points of view (12), and the like certain positions (13) as the images of the viewpoints (12) store.

14. The device according to claim 13, characterized in that it is arranged to perform the following steps for automatically imaging the viewpoints (12) on the corresponding positions (13) in the reference image (11):

Identifying a plurality of characteristic features and locations of these features in the reference image (11) by means of a feature detection algorithm, Identifying these features and determining locations of these features in the respective scene image (10) by applying the same or a corresponding algorithm to the scene image (10),

Determining a transformation that maps the locations of the features in the scene image (10) to the thus identified locations of the respective features in the reference image (11), and

Applying this transformation to the viewpoint (12) identified in the respective scene image (10).

15. Device according to one of claims 9 to 12, characterized in that it is arranged for each of the selected times, the corresponding position (13) in the reference image (11) together with the said these positions (13) associated with each attribute depending on a store the position (13) defining and associating it with the time or the scene image (10) associated with that time.

16. Device according to one of claims 9 to 15, characterized in that it is set up for a selected time points as defining such input or for an automatic selection of these times by applying one of these times as a function of the intervals (- 1 ₆ ) defining provision.

17. Device according to one of claims 9 to 16, characterized in that it further comprises an eye tracker (1) for detecting the viewing directions with a camera (4) for recording the scene images (10) and / or a camera (6) for Recording the reference image (11).